Skip to content

ARCHITECTURE.md

Latest commit

 

History

History
131 lines (100 loc) · 3.33 KB

ARCHITECTURE.md

File metadata and controls

131 lines (100 loc) · 3.33 KB

Architecture Documentation

System Overview

The AI Trading & Research Copilot is built as a modern web application with a clear separation between frontend, backend, and data layers.

Technology Stack

Backend

  • Framework: FastAPI (Python)
  • Database: PostgreSQL (relational data) + ChromaDB (vector embeddings)
  • Data Sources: yfinance, SEC EDGAR API, News APIs
  • AI/ML: OpenAI API, LangChain, Sentence Transformers

Frontend

  • Framework: Next.js 14 (React)
  • Charting: TradingView Lightweight Charts
  • Styling: Tailwind CSS
  • Language: TypeScript

Architecture Layers

1. Data Ingestion Layer

  • Scheduled ETL jobs for market data
  • News and filings ingestion pipelines
  • Data validation and cleaning

2. Storage Layer

  • PostgreSQL: OHLCV data, symbols, watchlists, user data
  • ChromaDB: Vector embeddings for RAG
  • Object Storage: Raw filings and documents (future)

3. API Layer

  • RESTful API endpoints
  • Authentication (future)
  • Rate limiting and caching

4. Business Logic Layer

  • Technical indicators computation
  • Risk scoring models
  • Backtest engine
  • Sentiment analysis

5. AI/ML Layer

  • RAG pipeline for document retrieval
  • LLM agent with tool calling
  • Embedding generation
  • Context construction

6. Frontend Layer

  • React components
  • Chart visualization
  • Chat interface
  • Dashboard and portfolio views

Data Flow

Market Data Flow

External API (yfinance) → DataFetcher Service → PostgreSQL → API Endpoint → Frontend

RAG Flow (Phase 2)

Documents → Chunking → Embedding → Vector DB → Retrieval → LLM → Response

Chat Flow (Phase 2)

User Message → API → RAG Retrieval → Context Construction → LLM → Response + Citations

Database Schema

Symbols Table

  • id, symbol, name, sector, industry, market_cap

Price Bars Table

  • id, symbol_id, timestamp, open, high, low, close, volume

Watchlists Table

  • id, user_id, symbol_id, created_at

API Endpoints

Phase 1

  • GET /api/v1/symbols - List symbols
  • GET /api/v1/symbols/{symbol} - Get symbol details
  • POST /api/v1/symbols/{symbol}/sync - Sync symbol data
  • GET /api/v1/prices/{symbol} - Get price bars
  • POST /api/v1/prices/{symbol}/sync - Sync price data
  • GET /api/v1/indicators/{symbol} - Get technical indicators

Phase 2 (Planned)

  • POST /api/v1/chat - AI copilot chat
  • GET /api/v1/news/{symbol} - Get news articles
  • GET /api/v1/filings/{symbol} - Get SEC filings

Phase 3 (Planned)

  • GET /api/v1/watchlist - Get user watchlist
  • POST /api/v1/watchlist - Add to watchlist
  • GET /api/v1/risk/{symbol} - Get risk metrics
  • GET /api/v1/risk/portfolio - Get portfolio risk

Phase 4 (Planned)

  • POST /api/v1/backtest - Run backtest
  • GET /api/v1/strategies - List strategy templates

Security Considerations

  • API authentication (JWT tokens)
  • Rate limiting
  • Input validation
  • SQL injection prevention (SQLAlchemy ORM)
  • CORS configuration

Scalability Considerations

  • Database indexing on frequently queried columns
  • Caching layer for price data (Redis - future)
  • Async processing for data ingestion
  • Horizontal scaling of API servers
  • CDN for static assets

Monitoring & Logging

  • Request/response logging
  • Error tracking
  • Performance metrics
  • Data freshness monitoring
  • AI model performance evaluation